An Open-Source Watertight Unmanned AerialVehicle for Water Quality Monitoring

Paulo Rodrigues, Francisco Marques, Eduardo Pinto, Ricardo Pombeiro,Andre Lourenco, Ricardo Mendonca, Pedro Santana, and Jose Barata

Abstract—This paper presents an open-source watertight mul-

tirotor Unmanned Aerial Vehicle (UAV) capable of vertical take-

off and landing on both solid terrain and water for environmental

monitoring. The UAV’s propulsion system has been designed so

as to also enable the active control of the UAV’s drift along the

water surface. This low power locomotion method, novel to such

a vehicle, aims to extend the available operation time on water

bodies’ surveys. The ability to take-off from water allows the

UAV to overcome any obstruction that appears on its path, such

as a dam. A set of field trials show the UAV’s water-tightness,

its take-off and landing capabilities in both land and water, and

also the ability to actively control its on-surface drifting.

I. INTRODUCTION

Current techniques for water monitoring rely on mannedteams [1], on remote information, such as the one provided bysatellites [2], [3], or on the deployment of Unmanned SurfaceVehicles (USV) [4], [5], [6], [7]. The remoteness, and vastnessof water bodies render the use of manned teams expensive anddangerous. Remote sensing offers an alternative but the datagets too often outdated, a fact that, added to its inadequacyto retrieve samples, significantly limits its applicability. Con-versely, a USV allows long-lasting monitoring of remote andvast water environments and also to retrieve water samples.

Despite relevant for environmental monitoring of waterbodies, there are many scenarios to which the use of a USV israther limited. For instance, the operation of such vehicles onopen sea may be severely compromised in face of powerfulwaves and strong tides. Such harsh conditions can damagesensory equipment or even the vehicles’ overall structure.On coastal regions, the risks are more related to the shore’sproximity, which induces turbulent waters and potentiallyreveal sharp-edged rocks. Sparsely distributed water bodies,either near the shore or inland (e.g., groups of small lakes orpuddles), posit a different challenge. Gathering data using aUSV in these regions is logistically cumbersome as it requireshuman intervention to carry the vehicle over land, renderingthe approach impractical. The same difficulties arise whennavigating to hard-to-reach locations due to the presence ofobstacles along the water bodies’ courses, such as dams,rapids, waterfalls, wave breakers, and debris (see Fig. 2). Aputative solution to this problem, yet to be addressed, is the use

P. Santana is with ISCTE - Instituto Universitario de Lisboa (ISCTE-IUL), Portugal and Instituto de Telecomunicacoes, Portugal. E-mail:pedro.santana@iscte.pt

P. Rodrigues, F. Marques, E. Pinto, R. Pombeiro, A. Lourenco,R. Mendonca, and J. Barata are with CTS-UNINOVA, UniversidadeNova de Lisboa (UNL), Portugal. E-mails: {pmr, fam, emp, rjp,

afl,rmm, jab}@uninova.pt

Fig. 1. The proposed UAV prototype taking-off from the water surface.

of amphibious vehicles that could autonomously navigate onland so as to reach the next water body’s segment. Althoughamphibious robots are already available [8], [9], the abilityto move them autonomously on unpredictable and unknownground is still an open problem. If such a vehicle wouldhave to overcome man-made structures, such as a dam, wouldmost probably require human intervention. Moreover, such anoperation would be excessively consuming in both time andenergy.

Waterways exploration using aerial vehicles has also beensubject of interest in recent years with efforts made forautonomous river exploration [10]. For in-situ water qualitymonitoring and sampling, rotary wing solutions [11], [12] havealso been proposed. In addition to be able to perform in-situwater sampling and retrieval, a watertight aerial vehicle canalways opt for the safest medium, either water, air, or landin case of an emergency (e.g., strong currents, rotor failure,strong winds).

An Unmanned Aerial Vehicle (UAV) performing environ-mental monitoring needs to be able to search for a suitablelanding site once its mission is concluded. This is a chal-lenging problem as potential dry landing areas on riverineenvironments are often either occluded by dense vegetation,such as tree canopy or low-hanging branches, or very difficultto distinguish from unreliable terrain on marshes and mudflats.Large waterways present a different challenge, namely, thelack of nearby landing areas, which requires a larger energeticautonomy from the aerial vehicle or a limited operation range.

An alternative to landing on dry ground is to allow the aerialvehicle to land on an autonomous surface vehicle also engaged

(a) (b)

Fig. 2. Two typical obstacles that a versatile autonomous vehicle should beable to tackle when performing environmental monitoring on water bodies.(a) A rapid in a riverine environment. (b) A wave breaker.

Fig. 3. A 3-D rendering of the proposed watertight UAV overcoming a damtowards the next water sampling waypoint, which would be unreachable forwater surface vehicles.

on the monitoring mission [6]. However, as already pointedout, relying on a USV may be limited in several scenarios.To relieve the UAV from this constant dependency, it needsto be able to land on the water surface itself. Bearing this inmind, this paper proposes an open-source1 autonomous andwatertight UAV with Vertical Take-Off & Landing (VTOL)capabilities on both land and water (see Fig. 1). The proposedUAV’s propulsion system has been designed to also allow theactive control of the UAV’s drift along the water surface. Thislow power and novel locomotion method aims at enabling longlasting water bodies’ surveys. In a sense, the UAV performs asa surface vehicle until an obstruction appears along its path.The obstruction is then overcome by the UAV by means of atake-off-flyby-landing sequence (see Fig. 3).

There are a few parallel developments on commercial re-motely controlled VTOL vehicles capable of landing on watersurfaces (refer to [13] for a recent survey). Differently fromthese parallel developments, the workload of the VTOL hereinpresented includes enough computational and sensory powerto enable autonomous behaviour. In fact, it is fully compliantwith the Robot Operating System (ROS) [14], which is thecurrent de facto standard in the robotics community. As aresult, the VTOL is easily integrated in larger robotic teamsand can also be easily extended with novel and advanced nav-igation and perception modules. Furthermore, the commercialnature of the parallel developments result in a closed designand absence of published systematic field trials. In addition to

1Construction diagrams, source code, and videos can be downloaded at theRIVERWATCH experiment site: http://riverwatchws.cloudapp.net/.

open the design to the community, we also report uncommonbehaviours for a watertight VTOL, such as controlled driftingon the water surface and low altitude flight exploiting theground effect. Finally, we provide the first specification of howlong-lasting environmental monitoring surveys in segmentedwater bodies could exploit the benefits of watertight VTOL.

This paper is organised as follows. Section II describesthe system’s design. Then, in Section III, the navigation andcontrol systems are presented. The results obtained in a setof field trials are summarized in Section IV. Finally, someconclusions and future work avenues are drawn in Section V.

II. SYSTEM DESIGN

A. Structure DesignThe watertight system’s main purpose is in-situ water mon-

itoring and, for that, it needs to fulfil a set of functionalrequirements. The robot should be able to take-off and landon solid ground as well as on water surfaces, while carryinga dedicated sensory payload for water sampling.

Fixed-wing aircraft have significantly more payload capa-bilities and energetic autonomy than rotary-wing vehicles.However, the extensive landing areas required by fixed-wingaircraft render a rotary-wing configuration with vertical take-off and landing capabilities better suited for the task at hand.To increase the UAV’s maximum payload with an additionalredundancy-based fault tolerance, a six rotor configuration waschosen. This is key to ensure that the aerial vehicle is able tokeep flying towards a safe landing site whenever required.

The major concern of using a UAV in a water environmentis the vehicle’s ability to sustain itself in the water surface.This demands for a watertight vehicle with enough buoyancyas to avoid damaging its control and propulsion systems. Toensure these properties, the vehicle core is made of carbonfibre with a high strength-to-weight ratio and low density corefilling. The UAV’s centre of buoyancy was designed so thatwhile the vehicle rests/floats on the water the propellers remainabove its surface, easing considerably any take-off procedure.In addition to buoyancy, the UAV needs to be robust againstcollisions with vegetation, other small obstacles, and the watersurface itself. This constraint is met with an enclosed propellerdesign (see Fig. 1). An additional feature available on the UAVis the ability to use its propellers’ thrust to cruise along thewater surface, using less energy than it would by flying. Thisnovel locomotion method allows the vehicle to exploit existingwater currents to expand its operational capacity followingriver-streams and leapfrogging obstacles. While being pulledby water currents, the vehicle needs only to perform smallmotion adjustments through the same propulsion system it usesto fly.

An important factor in real life monitoring campaigns isto ensure ease of use and maintenance. This includes astraightforward exchange of batteries and damaged electroniccomponents. To do so, an easily removable single central coverwas designed which relied on a screw-on top with an non-adhesive sealant material between the cover and the UAV’sframe. From the sealants tested, namely, silicone strips, rubber

(a) (b)

(c) (d)

Fig. 4. The UAV design process. (a) The structure design in SolidWorks. (b)The negative mold machined on a CNC. (c) The carbon sheet applied overthe mold. (d) The filling with microspheres

strips, and neoprene strips, the latter achieved the best results.When compressed tightly together, the neoprene strips ensuredwatertightness.

During the design and construction phases, a 3-D modelof the UAV was designed and validated with stress tests inSolidWorks. This design was later used to create a negativemold for the symmetrical top and bottom halves made ofCNC milled Polyurethane high density foam. The mold wasthen reinforced with several layers of microspheres and epoxyresin and highly polished. The bottom and top halves weremade individually with a low weight (90 g/m2) carbon fibrethat was vacuum molded, and cured at ambient temperature.To achieve the strongest and lightest structure possible, eachpart was filled with an innovative mixture of small spheresof polystyrene with Glass Bubbles (Microspheres) and epoxyresin. These allow the structure to resist the stress of successivelanding, take-off, and immersion procedures. Finally the twoparts were joined using carbon fibre tape impregnated withepoxy (see Fig. 4).

B. Hardware Architecture

The UAV’s propulsion is provided by six Altigator A3536motors equipped with 13 inches diameter and 4.5 inches pitchcarbon propellers, each capable of providing 2 kg of thrust.The ESCs (Electronic Speed Controllers) adopted were thehigh efficient 30A Afro ESCs with a firmware capable of1 kHz refresh rate. For energy supply, a single MaxAmps12000mAh XL 4s 120C LiPo battery or two 11000mAh 4s40C LiPo batteries provide between 15 minutes to 25 minutesof flight time, depending on the chosen battery configuration.

To support the low-level control system and high levelfunctionality, the UAV uses a modular architecture (see Fig. 7).Low-level and high-level boards are separated ensuring thatthe basic control is decoupled from functions supporting

Fig. 5. Hardware System Architecture overview.

(a) (b)

Fig. 6. Video acquisition system. (a) Gimbal placed inside the UAV structurewith a plastic dome. (b) Underwater images retrieved using the UAV’smonocular camera.

autonomous behaviour. The UAV is equipped with a VRBrainfrom Virtual Robotix for low-level control, which uses aGlobal Position System (GPS) device from Ublox, an onboardInertial Measurement Unit (IMU) based on MPU6500 anda MS5611 barometer for pose estimation. This low-levelboard is connected to an Odroid-XU from Hardkernell, whichprovides high-level processing capabilities, equipped with anExynos Octa-core CPU running the Indigo distribution ofRobot Operating System (ROS) [14], over a lightweight Linuxdistribution, namely the Xubuntu 14.04.

Perception is ensured by a downwards-facing RGB monocu-lar camera, placed on an active gimbal. The system is protectedinside a plastic dome for an unobstructed 150 degree view (seeFig. 6). Then, once the UAV floats on the water surface, andthe dome becomes completely submerged, underwater imagesmay now be captured.

III. NAVIGATION AND CONTROL

The software architecture was envisioned to improve theUAV’s robustness in harsh environments by decoupling high-level and low-level functionalities (see Fig. 7). This allowsthe UAV to revert to essential functionalities when it needsto save energy. This fallback is especially relevant in remoteenvironments, where the UAV will need to keep his energyexpenditure to a minimum. In this section an overview of theselected architecture is laid out.

Fig. 7. Software System Architecture overview.

The low-level control and basic navigation functions aresupported by a customized open source platform, the Ar-ducopter. This framework encompasses a flight controller withfeatures that range from autonomous take-off, landing, andwaypoint following. The communication between the low levelflight controller and higher level processing unit is establishedthrough MAVLink, which is a lightweight protocol designedfor aerial vehicles. A dedicated software module abstractsthe MAVLink protocol to comply with the Robot OperatingSystem (ROS) framework [14]. This allows seamless dataexchange between the flight controller and the higher levelnavigation ROS-based modules.

The sampling process is executed following a line-sweeppattern around the area the operator defines upon the UAV’saerial imagery. The UAV is then assigned with a given survey,involving one or several water bodies, to which it plans accord-ingly the most suitable path and mode of locomotion. Whileapproaching a water body, the UAV flies to the designatedinitial waypoint and lands on it once it is reached. There,the UAV changes its locomotion method so as to cruise onthe water surface through the several subsequent waypoints.While cruising, the UAV stores a dense data stream from itson-board payload sensors. Whenever a portion of land, or anyother obstruction (e.g., a dam), separates the UAV from itsnext waypoint, the UAV takes-off the water surface to fly overthe obstacle, and eventually land again on the next samplingpoint.

IV. FIELD TESTS

To validate the developed UAV, a set of experiments wereperformed. The first experiment was carried out on a controlledenvironment, namely, a swimming pool. Then, a few field trialswere run in a real world 85.000m2 water body.

A. Pool Tests

The objective of this first set of experiments was to assessthe UAV’s design appropriateness to a water environment.The trials began by checking the vehicle’s watertightness bysubmerging it at several depths, followed by buoyancy and

(a) (b)

Fig. 8. Preliminary validation in a swimming pool. (a) The UAV fullysubmerged. (b) The UAV hovering above the water surface.

rotational stability tests. Next, the platform was subjected todrop tests from different heights to check if it could survive acrash on the water surface without getting its inner componentsdamaged. Finally, take-off, landing, and on-surface locomotiontests were conducted.

To check its watertightness, the vehicle was submerged forfive minutes at depths 0.0m (fully submerged), 0.2m, 0.6m,and 1.0m (see Fig. 8(a)). Although it is not expected the UAVto be subjected to such extreme depths in real life applications,the robust operation under these circumstances assures thevehicle’s watertightness.

The UAV also has to be able to keep floating on the watersurface. Bearing this in mind, buoyancy tests were performedby loading the UAV with weights up to the double of its normaloperating payload of 1800 g. The relationship between thebuoyancy and gravity centres affects the stability of a vessel.To test this characteristic, the UAV was placed on the watersurface loaded with weights and rocked to see if it wouldtopple. The UAV remained stable during the tests.

To assess the vehicle’s endurance to the impact on thewater surface during an emergency landing, the UAV wasdropped from different heights, ranging from 0.5m up to1.5m with 0.25m steps. Repeated with pitch angles of 0degrees and 45 degrees, the UAV fell onto the water surfaceneither compromising its watertightness and buoyancy, neitherdamaging its frame, its propellers and their supports, or eventhe dome underneath.

Once the watertightness and robustness of the UAV wereassured, the different locomotion modes were tested. Take-offand landing tests consisted on placing the UAV on the watersurface, performing a take-off, hovering 1m above the water,and landing back on the same spot (see Fig. 8). An operatorrepeated the procedure five times, having the UAV taking-offwith ease and hover above the water surface, and landing backagain smoothly without any damage to the vehicle. Next, theon-surface locomotion was tested by remotely controlling theUAV across the pool length using its own propulsion system.This test was carried out with and without motion opposingwater currents induced by the pool’s water injection system.In both situations the vehicle navigated smoothly and withoutany water flooding the electronics compartment.

(a) (b)

Fig. 9. The site of the field trials. (a) Satellite image of the testing area (datedfrom 2002). (b) Aerial image retrieved by the UAV during the field trials.

(a) (b)

(c) (d)

Fig. 10. Sequence of snapshots taken during a take-off manoeuvre.

B. Field TrialsThe real world field trials were performed on a lake in

Sesimbra, Portugal (see Fig. 9). The objective was to assert thecapability of the UAV to operate in a sampling environment.The take-off/landing sequences performed on the pool weresuccessfully repeated in the field trial, which shows that theprocedure is robust enough for real world operation (seeFig. 10).

To be useful in real missions, the UAV needs not only tolift and land on the water surface but also to navigate in theenvironment and avoid obstacles. In this sense, longer flighttests were performed with the vehicle successfully flying fromone area of the lake to another while taking-off and landingon the water’s surface so as to overcome land present in thesampling path.

The on-surface locomotion method was also validated (seeFig. 11). The platform was remotely controlled to land on thewater surface and then to move along a line of 15m, driftingalong the lake surface. This type of locomotion showed to befeasible.

The use of ground effect for efficient flying near the surface

(a) (b)

(c) (d)

Fig. 11. Sequence of snapshots taken while drifting on the water surface.

(a) (b)

(c) (d)

Fig. 12. Sequence of snapshots taken while moving with ground effect.

was also tested. Ground effect in fixed wing aerial vehicles isfelt while flying lower that half the wingspan of the aircraft[15]. However, despite the UAV’s 0.3m propellers, the groundeffect on multirotors is present even at higher altitudes [16]. Inthe field trials, the UAV was able to exploit the ground effectto fly with reduced throttle input, suggesting an energy savingin the long run (see Fig. 12).

The water sealing of the UAV caused an expected rise intemperature that could affect the UAV reliability during along lasting survey. Still, according to the flight controller’stemperature log (see Fig. 13), no critical levels have beenreached, thus validating the design. If such a critical levelwould be reached, then the UAV would engage on a landingprocedure for cooling. The lowering of the temperature as aresult of resting on the water surface was actually observedthroughout the tests (see Fig. 13).

Finally, the chosen location for the field trials unveiledan interesting benefit of using watertight UAV’s for in-situwater sampling in detriment of surface vehicles. The satelliteimagery of this location (see Fig. 9(a)) shows a single waterbody to sample, whereas the imagery collected with the UAV’sonboard camera shows that a sand bank now splits the water

Fig. 13. UAV’s electronic case temperature evolution while engaging differentlocomotion modes. The blue segment corresponds to the moment in whichthe UAV was resting on the water surface.

body impairing a surface vessel to sample the entire location(see Fig. 9(b)). By using the updated image the watertightUAV could plan a way to avoid the obstacle by leapfroggingfrom one water body to the other.

V. CONCLUSIONS

To enlarge the scope of environmental monitoring roboticsin ecologically sensitive areas, an open-source watertightmultirotor aerial vehicle capable of take-off and landing onboth land and water was presented. The ability to operatein aquatic environments was attained by developing from theground up a watertight structure with significant buoyancy androbustness to low-altitude crashes against the water surface.These properties result from using a carbon fibre sealed framecombined with low density core filling. In addition to themechanical design, this paper also presented the hardwareand software architectures responsible for ensuring take-off,landing, and waypoint navigation. To ease the expansion ofthe vehicle’s capabilities and its integration onto larger roboticsystems, the software architecture is built upon the ROSframework.

A set of field trials confirmed the watertightness and ro-bustness of the structure to crashes against the water surface.Moreover, the tests also showed that the vehicle is able tosmoothly take-off, land and, especially, move along the watersurface by means of the same propulsion system used forflying. The experiments further confirmed the ground effect tobe an efficient and controlled alternative locomotion method.The ability to land on water is key to handle emergencylandings. The ability to move on the water surface enableslong lasting water sampling by exploiting water currents. Theability to take-off and landing on water allows the vehicle tofly over obstacles, such as a dam or a large portion of land.

As future work, we expect to develop a set of autonomousbehaviours capable of exploiting the water/land navigationcapabilities of the presented vehicle towards energy efficientmanagement. This includes controlling the aspects of eachlocomotion mode (e.g., motion on the water surface), deter-

mining the most suited locomotion method for a given context,and optimise the waypoints distribution to best exploit windsand currents when drifting on the water surface.

We also expect to fuse the abilities of the developedsystem with the long-lasting characteristics of the RiverwatchExperiment [6], allowing persistent monitoring of ecologicalrelevant areas. A complete water sampling campaign is beingprepared to further assess the accuracy and robustness of thesystem.

ACKNOWLEDGMENTS

This work was partially funded by the CTS multi-annualfunding, through the PIDDAC Program funds.

REFERENCES

[1] C. Flener, M. Vaaja, A. Jaakkola, A. Krooks, H. Kaartinen, A. Kukko,E. Kasvi, H. Hyyppa, J. Hyyppa, and P. Alho, “Seamless mappingof river channels at high resolution using mobile liDAR and UAV-photography,” Remote Sensing, vol. 5, no. 12, pp. 6382–6407, 2013.

[2] L. C. Smith, “Satellite remote sensing of river inundation area, stage,and discharge: A review,” Hydrological processes, vol. 11, no. 10, pp.1427–1439, 1997.

[3] M. Fingas and C. Brown, “Review of oil spill remote sensing,” MarinePollution Bulletin, vol. 83, no. 1, pp. 9–23, 2014.

[4] M. Dunbabin, A. Grinham, and J. Udy, “An autonomous surface vehiclefor water quality monitoring,” in Proc. of Australasian Conference onRobotics and Automation 2009, 2009, pp. 1–6.

[5] J. Leedekerken, M. Fallon, and J. Leonard, “Mapping complex marineenvironments with autonomous surface craft,” in Proc. of the 12thInternational Symposium on Experimental Robotics, 2010, pp. 1–15.

[6] E. Pinto, F. Marques, R. Mendonca, A. Lourenco, P. Santana, andJ. Barata, “An autonomous surface-aerial marsupial robotic team forriverine environmental monitoring: Benefiting from coordinated aerial,underwater, and surface level perception,” in Proc. of the IEEE Inter-national Conference on Robotics and Biomimetics (ROBIO). IEEE,2014.

[7] H. Ferreira, C. Almeida, a. Martins, J. Almeida, N. Dias, a. Dias,and E. Silva, “Autonomous bathymetry for risk assessment with ROAZrobotic surface vehicle,” in Proc. of OCEANS ’09 IEEE Bremen, 2009,pp. 1–6.

[8] U. Saranli, M. Buehler, and D. E. Koditschek, “Rhex: A simple andhighly mobile hexapod robot,” The International Journal of RoboticsResearch, vol. 20, no. 7, pp. 616–631, 2001.

[9] L. Cui, P. Cheong, R. Adams, and T. Johnson, “Ambot: A bio-inspiredamphibious robot for monitoring the swan-canning estuary system,”Journal of Mechanical Design, vol. 136, no. 11, p. 115001, 2014.

[10] S. Jain, S. Nuske, A. Chambers, L. Yoder, H. Cover, L. Chamberlain,S. Scherer, and S. Singh, “Autonomous river exploration,” in Field andService Robotics. Springer, 2015, pp. 93–106.

[11] J.-P. Ore, S. Elbaum, A. Burgin, B. Zhao, and C. Detweiler, “Au-tonomous aerial water sampling,” in Field and Service Robotics.Springer, 2015, pp. 137–151.

[12] M. Schwarzbach, M. Laiacker, M. Mulero-Pazmany, and K. Kondak,“Remote water sampling using flying robots,” in Proc. of InternationalConference on Unmanned Aircraft Systems (ICUAS), 2014. IEEE, 2014,pp. 72–76.

[13] X. Yang, T. Wang, J. Liang, G. Yao, and M. Liu, “Survey on the novelhybrid aquatic–aerial amphibious aircraft: Aquatic unmanned aerialvehicle (aquauav),” Progress in Aerospace Sciences, vol. 74, pp. 131–151, 2015.

[14] M. Quigley, B. Gerkey, K. Conley, J. Faust, T. Foote, J. Leibs, E. Berger,R. Wheeler, and A. Ng, “Ros: an open-source robot operating system,”in Proc. of the ICRA Open-Source Software Workshop, 2009.

[15] K. Rozhdestvensky, “Wing-in-ground effect vehicles,” Progress inAerospace Sciences, vol. 42, p. 211283, 2006.

[16] S. Aich, C. Ahuja, T. Gupta, and P. Arulmozhivarman, “Analysis ofground effect on multi-rotors,” in Proc. of the International Confer-ence on Electronics, Communication and Computational Engineering(ICECCE), 2014. IEEE, 2014, pp. 236–241.